Disclaimer

“But why not just use VSCode?”

Python install and package management

Setting up RStudio to run Python

Setting up RStudio to run Python

Enter the path to the python interpreter in your base environment

Setting up RStudio to run Python

2. Install the reticulate package

install.packages("reticulate")

What does reticulate do?⁵

A note from Posit about the philosophy behind Python tools in Rstudio

These tools are not intended for standalone Python work but rather explicitly aimed at the integration of Python into R projects (and as such are closely tied to the reticulate package).

They “strongly suggest” using one of the IDEs available for doing data science in Python for Python-only projects.⁶

What does reticulate do?

3. Provides flexible binding to different versions of Python including virtual environments and Conda environments

Set your environment

First thing you always want to do is set your environment:

library(reticulate)

Warning: package 'reticulate' was built under R version 4.4.1

use_condaenv("iffrug")

What does reticulate do?

1. Allows calling Python from R in multiple ways including from RMarkdown/Quarto, sourcing Python scripts, importing Python modules, and using Python interactively within an R session

The py object

When you call library(reticulate), it creates the py object in the reticulate package environment.

Importing modules

reticulate::import() can be used to import any installed Python module into your R environment.

os <- import("os")

os$listdir(".") |> head()

[1] "teal-deer.jpg"                          
[2] "2024-04-24_python-in-rmarkdown.Rmd"     
[3] "2024-04-29_IFFRUG_python-in-rstudio.qmd"
[4] "add.py"                                 
[5] "arrested-development-david-cross.gif"   
[6] "diva-girl.gif"                          

builtins <- import_builtins()
builtins$print('Hello, World!')

Hello, World!

Sourcing scripts

Let’s say I have a Python script that defines a function:

def add(x,y):
  return x + y

If I’d like to use that function in R, I can source it using source_python().

source_python('add.py')
add(5,10)

[1] 15

Executing code

Let’s say my collaborator wrote a Python script for processing some raw data. I’d like to work with the processed data in R, but my collaborator only provided me with the raw data and the script.

I can use py_run_string() and py_run_file() to process the data, and then access any objects created into the Python main module using the py object exported by reticulate:

# Set the Python variable pointing to the raw data file
py_run_string("file = 'rawdata.csv'")

# run the processing script, which takes the file argument
py_run_file("process_raw_data.py")

# access the resulting df
py$df

  a b
1 1 4
2 2 5
3 3 6

Working in RMarkdown⁷

reticulate includes a Python engine for RMarkdown, and knitr v 1.18 and higher uses this engine by default.

Then you can start inserting Python chunks just like you would R chunks, and knitr will knit everything together:

Just like when working interactively, you can access objects created in Python chunks in R by using the py object:

And you can access objects created in R chunks in Python by using the r object:

Working in Quarto

Quarto provides all the support for Python that RMarkdown does, plus support for Jupyter:

• You can also provide a full kernelspec in your YAML:

---
title: "My Document"
jupyter:
  kernelspec:
    name: xpython
    language: "python"
    display_name: "Python 3.7 (XPython)"
---

quarto render document.ipynb

What does reticulate do?

2. Translates between R and Python objects (e.g., between R and Pandas data frames, or between R matrices and NumPy arrays)

Type conversions

When calling into Python, R data types are automatically converted to their equivalent Python types.

More on object conversion

The automatic conversion between R types and Python types works well in most cases, but sometimes you might want more control over the conversions.

Controling when conversion happens

If you’d like to work directly with Python objects by default you can pass convert = FALSE to the import() function.

# import numpy and specify no automatic Python to R conversion
np <- import("numpy", convert = FALSE)

# do some array manipulations with NumPy
a <- np$array(c(1:4))
(sum_np <- a$cumsum())

array([ 1,  3,  6, 10])

# what is sum?
class(sum_np)

[1] "numpy.ndarray"         "python.builtin.object"

Then when you’re done working with the object in Python, you can convert it to an R object explicitly with py_to_r().

# convert to R explicitly at the end and print object
(sum_r <- py_to_r(sum_np))

[1]  1  3  6 10

# what is sum_r?
class(sum_r)

[1] "array"

Defining the conversion

Numeric types are different between R and Python. For example, 42 in R is a float, while in Python it’s an integer.

If you want to explicitly define a number as an integer in R so that it’s passed as such to Python, use the L suffix:

class(42)

[1] "numeric"

class(42L)

[1] "integer"

If a Python API requires a list but you’re only passing it a single element, you can wrap it in base list():

foo$bar(indexes = list(42L))

tuple("a", "b", "c")

dict(foo = "bar", index = 42L)

Indices

Python uses 0-based indices for collections:

sum_np[0L]

1

while R uses 1-based indices:

sum_r[1]

[1] 1

Python indices are non-inclusive for the end range, while R indices are:

sum_np[2L:4L]

array([ 6, 10])

sum_r[2:4]

[1]  3  6 10

sum_np[-1L]

10

sum_r[-1]

[1]  3  6 10

Important points about arrays¹⁰

R and Python represent arrays and matrices in memory differently:

• R only supports column-major order (FORTRAN-style)
• NumPy supports both column-major and row-major (C-style) order, but defaults to row-major

The most important thing to remember about this is that

R and Python print arrays differently.

(x <- np$arange(1, 9)$reshape(2L, 2L, 2L))

array([[[1., 2.],
        [3., 4.]],

       [[5., 6.],
        [7., 8.]]])

(y <- py_to_r(x))

, , 1

     [,1] [,2]
[1,]    1    3
[2,]    5    7

, , 2

     [,1] [,2]
[1,]    2    4
[2,]    6    8

Sparse matrices

reticulate supports the conversion of sparse matrices created by the Matrix R package to and from SciPy CSC matrices.¹¹

Important points on data frames

As mentioned earlier, R data frames can be automatically converted to and from Pandas data frames. By default, columns are converted using the same rules governing R array <=> NumPy array conversion, with a couple extensions:

• R factors <=> Python categorical variables
• R POSIXt times <=> NumPy array with dtype=datetime64[ns]

Important points on data frames

If the R data frame has row names, the generated Pandas data frame will be re-indexed using those row names, and vice versa.

Using Pandas nullable data types

Pandas out of the box handles NAs differently than R:

(df <- data.frame(
  int = c(NA, 1:4),
  num = c(NA, rnorm(4)),
  lgl = c(NA, rep(c(TRUE, FALSE), 2)),
  string = c(NA, letters[1:4])
))

  int        num   lgl string
1  NA         NA    NA   <NA>
2   1  1.2242601  TRUE      a
3   2 -0.1260265 FALSE      b
4   3  2.3851476  TRUE      c
5   4  1.5463105 FALSE      d

r_to_py(df)

          int       num    lgl string
0 -2147483648       NaN  False   None
1           1  1.224260   True      a
2           2 -0.126026  False      b
3           3  2.385148   True      c
4           4  1.546311  False      d

However, Pandas has experimental support for nullable data types (represented by pd.NA), but you have to enable it first:

# tell Pandas to use NAs
options(reticulate.pandas_use_nullable_dtypes = TRUE)

r_to_py(df)

    int       num    lgl string
0  <NA>      <NA>   <NA>   <NA>
1     1   1.22426   True      a
2     2 -0.126026  False      b
3     3  2.385148   True      c
4     4  1.546311  False      d

For advanced Python users

For advanced Python users, there’s more documentation on

• Contexts
• Iterators
• Functions
• Creating high-level R interfaces for Python libraries

Check out https://rstudio.github.io/reticulate/articles/calling_python.html for deets.

Access to Python help from R

You can print documentation on any Python object using py_help():

py_help(os$chdir)

This will open a text document outside of RStudio:

Help on built-in function chdir in module nt:

chdir(path)
    Change the current working directory to the specified path.

    path may always be specified as a string.
    On some platforms, path may also be specified as an open file descriptor.
      If this functionality is unavailable, using it raises an exception.

There is also this excellent article aimed at R users who are new to Python.

More on arrays

• Dense R arrays are presented to Python as column-major NumPy arrays (FORTRAN-style).
• All NumPy arrays (column-major, row-major, or otherwise) are presented to R as column-major arrays, since that’s all R can understand.
• R arrays are only copied to Python when they need to be, otherwise data are shared.
• NumPy arrays are always copied when moved into R arrays. This can sometimes lead to multiple copies of any one array in memory at one time.

Remember this 2x2x2 array?

(x <- np$arange(1, 9)$reshape(2L, 2L, 2L))

array([[[1., 2.],
        [3., 4.]],

       [[5., 6.],
        [7., 8.]]])

(y <- py_to_r(x))

, , 1

     [,1] [,2]
[1,]    1    3
[2,]    5    7

, , 2

     [,1] [,2]
[1,]    2    4
[2,]    6    8

Python groups by the first index when printing, while R groups by the last index:

x

array([[[1., 2.],
        [3., 4.]],

       [[5., 6.],
        [7., 8.]]])

x[0L,,]

array([[1., 2.],
       [3., 4.]])

x[,,0L]

array([[1., 3.],
       [5., 7.]])

y

, , 1

     [,1] [,2]
[1,]    1    3
[2,]    5    7

, , 2

     [,1] [,2]
[1,]    2    4
[2,]    6    8

y[1,,]

     [,1] [,2]
[1,]    1    2
[2,]    3    4

y[,,1]

     [,1] [,2]
[1,]    1    3
[2,]    5    7

What about arrays from R to Python?

In the previous example I created an array in Python and ported it to R. What about the other way around?

(v <- array(1:24, c(4, 3, 2)))

, , 1

     [,1] [,2] [,3]
[1,]    1    5    9
[2,]    2    6   10
[3,]    3    7   11
[4,]    4    8   12

, , 2

     [,1] [,2] [,3]
[1,]   13   17   21
[2,]   14   18   22
[3,]   15   19   23
[4,]   16   20   24

(w <- np$array(v))

array([[[ 1, 13],
        [ 5, 17],
        [ 9, 21]],

       [[ 2, 14],
        [ 6, 18],
        [10, 22]],

       [[ 3, 15],
        [ 7, 19],
        [11, 23]],

       [[ 4, 16],
        [ 8, 20],
        [12, 24]]])

The NumPy array will be created using column-major ordering:

w$flags

  C_CONTIGUOUS : False
  F_CONTIGUOUS : True
  OWNDATA : True
  WRITEABLE : True
  ALIGNED : True
  WRITEBACKIFCOPY : False

Remember:

• F for “FORTRAN” (column-major order)
• C for “C” (row-major order)

You can always create NumPy arrays in column-major order by passing the "F" flag:

np$reshape(np$arange(1, 25), c(4L, 3L, 2L))

array([[[ 1.,  2.],
        [ 3.,  4.],
        [ 5.,  6.]],

       [[ 7.,  8.],
        [ 9., 10.],
        [11., 12.]],

       [[13., 14.],
        [15., 16.],
        [17., 18.]],

       [[19., 20.],
        [21., 22.],
        [23., 24.]]])

np$reshape(np$arange(1, 25), c(4L, 3L, 2L), "F")

array([[[ 1., 13.],
        [ 5., 17.],
        [ 9., 21.]],

       [[ 2., 14.],
        [ 6., 18.],
        [10., 22.]],

       [[ 3., 15.],
        [ 7., 19.],
        [11., 23.]],

       [[ 4., 16.],
        [ 8., 20.],
        [12., 24.]]])

Reshaping arrays

The R dim() function is used to reshape arrays in R. This works by changing the dim attribute of the array, effectively re-interpreting the array indices using column-major semantics.

# NumPy reshape uses row-major semantics
np$reshape(np$arange(1,5), c(2L,2L))

array([[1., 2.],
       [3., 4.]])

To overcome this, use reticulate::array_reshape() to reshape R arrays using row-major semantics.

# make an array from a vector
# of 4 elements:
u <- 1:4

# dim() uses
# column-major semantics
dim(u) <- c(2,2)
u

     [,1] [,2]
[1,]    1    3
[2,]    2    4

# array_reshape() uses
# row-major semantics
array_reshape(1:4, c(2,2))

     [,1] [,2]
[1,]    1    2
[2,]    3    4